topic Re: Combining Imputations in Proc PHREG after Multiple Imputation in Statistical Procedures

Combining Imputations in Proc PHREG after Multiple Imputation

wbaker0621 — Thu, 11 Jun 2020 11:23:25 GMT

Greetings All, I am a non-statistician and have only been using SAS for a few months. I am working with a medical registry to conduct survival analysis and have observed missing data at random. I have run Proc MI to create 5 imputation data sets, but am getting stuck on the 2nd step when I analyze the set using Proc PHREG. Examples I have seen call for outputing the parameter estimates using the ODS Output statement (see below). When I then attempt to sort that new dataset prior to using Proc MIANALYZE it does not seem to exist. I would appreciate any advise on this code and examples of successful code for these 3 steps (Proc MI, Proc PHREG, Proc MIANALYZE). Thank you so much for your help in advance.

Cheers

Bill

/* Use multiple imputation for missing data values */
proc mi data=SRTR.BLACK_TX2 round=0.001 nimpute=5 seed=123456 out=SRTR.BLACK_MI noprint;
	class &VARLIST_CAT;
	var &VARLIST_CAT &VARLIST_CONT;
	fcs;
run;

/* Run Cox regression of multiply imputed data */
proc phreg data=SRTR.BLACK_MI;
	by _Imputation_;
	class Group &varlist_cat;
	model TTE_10 * Death_10(0) = Group &varlist_cat &varlist_cont
		/ ties=efron rl;
	ods output ParameterEstimates=SRTR_BLACK_ES;
run;

Re: Combining Imputations in Proc PHREG after Multiple Imputation

SteveDenham — Thu, 11 Jun 2020 12:32:59 GMT

What does it say in the log? Is SRTR_BLACK_ES created, or does something else happen? Without the log, it is very hard to tell what might be happening.

SteveDenham

Re: Combining Imputations in Proc PHREG after Multiple Imputation

SAS_Rob — Thu, 11 Jun 2020 13:00:36 GMT

It would be helpful to see what exactly isn't working in the MIANALYZE step with the LOG.

But the MIANALYZE code should just be something like this:

proc mianalyze parms(classvar=classval)=SRTR_BLACK_ES;

class Group &varlist_cat;
modeleffects Group &varlist_cat &varlist_cont;

run;

Re: Combining Imputations in Proc PHREG after Multiple Imputation

wbaker0621 — Thu, 11 Jun 2020 13:02:45 GMT

Great point. This newbie still hasn't gotten used to looking at the log. The datafile was created, but was put in the work folder.

Re: Combining Imputations in Proc PHREG after Multiple Imputation

wbaker0621 — Thu, 11 Jun 2020 13:42:29 GMT

Thanks all. This combination of code worked, once I payed attention to the log and knew where the output files were headed.

/* Run Cox regression of multiply imputed data */
proc phreg data=SRTR.BLACK_MI;
	by _Imputation_;
	class Group(ref='WHITE_WOMAN') &varlist_cat;
	model TTE_10 * Death_10(0) = Group &varlist_cat &varlist_cont
		/ ties=efron rl;
	ods output ParameterEstimates=SRTR_BLACK_ES;
run;

/* Combine model coefficients */
proc sort data=work.srtr_black_es;
	by Parameter ClassVal0 _Imputation_;
run;

proc mianalyze data=work.srtr_black_es;
	by Parameter ClassVal0;
	modeleffects Estimate;
	stderr StdErr;
	ods output ParameterEstimates = SRTR_Black_es_mianal;
run;

/* Exponentiate in order to obtain hazard ratio estimates and confidence intervals */
data srtr.black_HR; set work.srtr_black_es_mianal;
	Log_HR_comb=Estimate;
	HR_comb=exp(Estimate);
	HR_LCL_comb=exp(LCLMean);
	HR_UCL_comb=exp(UCLMean);
	keep Parameter ClassVal0 HR_comb HR_LCL_comb HR_UCL_comb Probt;
	rename Probt=HR_pval_comb;
run;